Adaboost Algorithm-based Face Detection Research

Face detection is a fundamental and important research theme in the topic of Pattern Recognition and Computer Vision, it has a broad application in many fields such as content-based image and video retrieval, video surveillance, automatic face recognition and human-computer interface, etc.Given an arbitrary image, the goal of face detection is to determine whether or not there are any faces in the image and, if present, return the image location and extent of each face. In this paper, the AdaBoost cascade face detection algorithm proposed by Viola et al. is analyzed in detail. The main contributions are as follows:1,This article analyzes the training error and the generalization error of AdaBoost algorithms systematically, and builds a face detection system while in-depth study on face detection method based on AdaBoost algorithm.2,This article redefines the training error which is caused when we train the weak classifier, and proposes MCE-AdaBoost algorithm. The new definition of training error will pay more attention to the error which erroneously estimates the face sample as non-face sample, this much more conforms to face detection of this special target detection issue. The experimental results show that MCE-AdaBoost algorithm can effectively improve the detection performance of the final classifier.3,Focusing on the disadvantages of AdaBoost algorithms: If difficult samples exist in the training samples, with the iterative number increasing, this easily leads to degeneration phenomenon, and reduces the generalization ability of the classifier. This article proposes LWE-AdaBoost algorithm which can limit weight expansion, the experimental results indicate that the LWE-AdaBoost algorithm can restrain the occurrence of degeneration phenomenon well.4,In the traditional AdaBoost method, the weights are updated based on the weak classifier's overall error rate, this article proposes one kind of the method which unifies the level of difference between the threshold and the feature value of the weak classifier with the weak classifier's overall error rate. Compared to the method which only based on the overall classification error rate to update the weights, this method can achieve higher detection rate while reduces the mistaken-detection rate.